Learning without Neural Networks

The conscious part of the mind favors learning by comparison.  It doesn't need much in the way of resources to do this - 10% seems like more than enough.

The unconscious part of the mind seems captured accurately enough with neural networks and training.  For the time being lets pretend it uses the other 90%.

From the research done over the past 40 years it's reasonable to claim that a vague awareness of many things is well represented by the neural network.  That accounts for 1/2 of a more mind like AI and leaves some questions about the rest.

Assuming someone hasn't already solved this problem and answered all questions...

How could we design it?

Well... how about along the line of consciously comparing 2 things?  Just simple comparison based learning where aggregations of awareness of few things are compared and related.

Terms - the reference used for the descriptions below is Theory

• Thinking Space - for the conscious sub mind it's where detailed awareness of a few things can be compared and related. awareness of new knowledge/answers can be gained through modeling, correlation, extrapolation, projection. all are implicit but can be made explicit by confirmation.
• Vague awareness - the simplest unit of awareness.  a vague awareness of something simple/finite/discrete.
• Unconscious awareness - a vague awareness of many things that are simple/finite/discrete.
• Deep awareness - a deep awareness of few things that are complex/large/abstract.
• Conscious awareness - the aggregation of unconscious awareness.  the 7 +- 2 rule is probably overstated here.  would guess it's more like 5 +- 1 due to the number of permutations increasing with each additional awareness added to the comparison.

Only 2 things are ever compared at once.  More than 2 things can be compared but it happens in pairs.  Should that require a more fine grained look then detailed awareness is broken down into a hierarchy of varying amounts of vague awareness until the specific level of detail is reached.  That's done with each thing being compared until they're both reduced to the correct (and same) level of awareness detail.

We model what we know.

When it's necessary to go beyond what a model supports we:

• Learn new knowledge
• Create new knowledge

The conscious sub mind uses Reason.

The unconscious sub mind uses Reflex.

Inference - a process of inferring something (knowledge) through reasoning

• Deductive reasoning (Deduction) - in this case taking 2 knowledge/answers as statements and reasoning a conclusion
• Inductive reasoning (Induction) - in this case knowledge/answers are synthesized to a general truth for a conclusion
• Abductive reasoning (Abduction) - inference to the best available conclusion.  does not verify the conclusion, at least not initially

Creating new knowledge/answers through Inference (Inductive reasoning):

• If thing A has detail X with some property
• And thing B has detail X
• Then thing B, detail X should have some property
• If some property is confirmed (and sometimes when it isn't) it can become a new knowledge/answer
• When a new knowledge/answer is correct it can be built on to create others
• Should a new knowledge/answer be wrong, all others based on it become wrong

New knowledge/answers can also be created by identifying a correlation:

    Once a correlation is identified then an extrapolation or interpolation can be created.

Informative link:  Lesson 12..8 - Extrapolation on PennState Eberly College of Science